AI labs are buying data from dead startups to train their models
When Keelan Miskell began winding down his crypto startup this year, he didn’t think its data was worth much: the users were anonymous, the company just a few years old. But he’s since turned the firm’s digital archives into real money.
A new industry has emerged to help failing startups monetize their remaining digital assets, including code, emails, chats, and documents. Companies like SimpleClosure and Sunset act as brokers, selling these assets to AI labs, which use them to train artificial intelligence models. For example, Keelan Miskell earned $25,000 for his startup's code and an estimated $15,000 for workspace data stored in platforms like Google, Notion, Figma, and Slack. SimpleClosure’s Asset Hub platform has facilitated over 150 deals, with projections to return over $10 million to startups by the end of 2026. This trend reflects a broader rise in startup shutdowns, with estimates showing a 26% to 56% annual increase in recent years.
AI training labs are increasingly turning to real-world data from defunct companies because public and synthetic datasets have limitations. For instance, Surge AI buys data to create reinforcement-learning gyms, where AI agents practice tasks like debugging IT systems or drafting legal documents. Real data archives, such as Slack conversations or email histories, are preferred over synthetic data because they reflect authentic workplace dynamics, including disagreements or judgment calls. However, the performance gains from using real data remain unclear, as no company has yet published comparative tests. Pricing for such data varies widely due to the lack of standardized valuation methods, with buyers relying on research estimates rather than proven metrics.
Large-scale data sales from defunct companies have already occurred, such as Google’s $10 million purchase of Spirit Airlines’ decades of data during its bankruptcy. This included 100 million emails, 500 million Microsoft Teams messages, and 176,000 employee records. However, legal and ethical concerns surround these transactions. Spirit Airlines’ flight attendants’ union objected on privacy grounds, and legal experts like Greg Ewing describe such data as legally radioactive due to the potential presence of sensitive personal details in emails and calendars. While brokers and buyers claim to scrub personally identifiable information, critics argue that proper licensing agreements and worker consent are often lacking.
The financial benefits of these data sales typically flow to senior creditors and investors, not employees, raising questions about fairness. Ifeoma Ajunwa, a law professor at Emory University, describes this as a form of *captured capital*, where workers’ data is sold without their explicit consent. Currently, no federal law prohibits such sales, though legal precedent suggests monitoring for work purposes ends when a company shuts down. Additionally, founders often lack transparency about who buys their data or its intended use, and there is little assurance that the sale price is fair. The market for defunct companies’ data is expected to stabilize in the coming year, though this depends on the continued growth of AI demand.
